Research On Optimization Calculation Methods Of Two Memory Hard Hash Function

With the rapid development of social economy and the rise of cloud computing,Internet services have brought convenience to people's lives.At the same time,information security has become increasingly important.The Memory hard hash function is difficult to implement efficiently on heterogeneous hardware platforms,and can protect people's information better.In this context,we'll make a further research on Memory hard hash function,and will propose a performance optimization of Memory hard hash function based on SW26010 processor and GPU in this context.On the SW26010 processor,we'll take the advantage of the large number of cores,adopt the master-slave parallel acceleration mode to optimize performance and make slave cores compute the hash function.To solve the problem that the local storage in the slave cores is not enough for hash function,we propose two solutions in this context that the one is storing all intermediate data in main memory,the another is compressing the intermediate data and storing in local memory.Then we optimize the solution that storing all intermediate data in main memory by hide the latency of slave core memory access,we propose two optimization schemes: master and slave core memory combined calculate multiple plaintexts and cross calculate multiple plaintexts only using master core memory,and optimizing them using SIMD instructions.Compared with the performance of the hash function before optimization,the performance of Scrypt and Argon2 d has been improved.On the GPU platform,we take batch computing to improve the parallelism of the program.Then implementing merge access memory by rearrange the data according to the GPU memory model to improve the performance of the hash function.Finally,for the parallelizable part of the hash function,we use multiple threads to compute the same plaintext together,and use CUDA shuffle instructions to accelerate the exchange of data between GPU threads.Compared with the performance before optimization the performance of Scrypt and Argon2 d has been effectively improved,and the performance of Scrypt is 2.03 times that of hashcat.